, ,

Motivated by the observation that for a compact spin Riemannian manifold one can recover its smooth structure, its Riemannian metric, and much else, directly from its standard Dirac operator, Connes pointed out that from a spectral triple

(A, ℋ, D)

one obtains a metric on the state space

S (A)

of the unital

C^{*}

-algebra

A

by the formula

ρ_{D} (φ, ψ) = sup {| φ (a) - ψ (a) | : ∥ [D, a] ∥ \leq 1, a \in A},

{a \in A; ∥ [D, a] ∥ \leq 1} / (C 1)

is bounded[4] .

A natural question is that when the topology on

S (A)

determined by

ρ_{D}

coincides with the

w^{*}

-topology. Rieffel has studied it in a more general situation in which

A

is just an order unit space and

∥ [D, \cdot] ∥

is replaced by a Lipschitz seminorm on

A

[9, 10, 11] . This and certain statements in the high energy physics and string theory, concerning non-commutative spaces that converged to other spaces, led him to the concept of compact quantum metric spaces[12, 13] .

For a discrete group

G

, which is of rapid decay with respect to some length function, Antonescu and Christensen got a metric on the state space

S (C_{r}^{*} (G))

of the reduced group

C^{*}

-algebra

C_{r}^{*} (G)

which is finite for all pairs, is bounded on

S (C_{r}^{*} (G))

, and generates the

w^{*}

-topology on

S (C_{r}^{*} (G))

[1] . This way of obtaining a metric from higher derivatives inspired them to discuss metric spaces without a smooth structure. They believe that any norm compact balanced convex subset of a unital

C^{*}

-algebra

A

which separates the states on

A

contains much information needed, and call the subset a metric set of

A

. In particular, they showed that it works well with respect to a translation of the classical Arzelà-Ascoli theorem into a non-commutative language[1] .

Most of interesting constructions in view of Lipschitz seminorms on

C^{*}

-algebras, such as those from Dirac operators, or those in [9] , also provide in a natural way seminorms on all the matrix algebras over the algebras. Rieffel suggested that some “matrix Lipschitz seminorm” in analogy with the matrix norms of [5] will be of importance[10] . In [15, 16] , we developed a version of it on the matrix order unit spaces. It has many nice properties[15, 16, 17] . In [1] , Antonescu and Christensen asked if their result on the non-commutative version of the Arzelà-Ascoli theorem is valid in a wider generality like operator systems. The main goal of this paper is to give an affirmative answer to their question both at the “matrix” level and at the “function” level.

This paper is organized as follows. We begin in Section 2 with a discussion of the notions. Because we need corresponding equicontinuity of mappings on the framework of operator spaces, we discuss the continuous matrix mappings on matrix metric spaces in Section 3 . We introduce the concept of matrix metric sets of a matrix order unit space

(A, 1)

which is closely related to matrix metrics on the matrix state space

C S (A)

generating the BW-topology. This is done in Section 4 .

We prove our main results (Theorem 5.4 and Corollary 5.5 ) on relative compactness in Section 5 .

2 Preliminaries

All vector spaces are assumed to be complex throughout this paper. Given a vector space

V

, we let

M_{m, n} (V)

denote the matrix space of all

m

n

matrices

v = [v_{i j}]

with

v_{i j} \in V

, and we set

M_{n} (V) = M_{n, n} (V)

. If

V = C

, we write

M_{m, n} = M_{m, n} (C)

and

M_{n} = M_{n, n} (C)

, which means that we may identify

M_{m, n} (V)

with the tensor product

M_{m, n} \otimes V

. We identify

M_{m, n}

with the normed space

ℬ (C^{n}, C^{m})

We use the standard matrix multiplication and *-operation for compatible scalar matrices, and

1_{n}

for the identity matrix in

M_{n}

, and

0_{m, n}

for the

m

n

zero matrix. There are two natural operations on the matrix spaces. For

v \in M_{m, n} (V)

and

w \in M_{p, q} (V)

, the direct sum

v \oplus w \in M_{m + p, n + q} (V)

is defined by letting

v \oplus w = [\begin{matrix} v & 0 \\ 0 & w \end{matrix}],

and if we are given

α \in M_{m, p}

v \in M_{p, q} (V)

and

β \in M_{q, n}

, the matrix product

α v β \in M_{m, n} (V)

is defined by

α v β = [\sum_{k, l} α_{i k} v_{k l} β_{l j}] .

A *-vector space

V

is a complex vector space together with a conjugate linear mapping

v \mapsto v^{*}

such that

v^{* *} = v

. A *-vector space

V

is said to be matrix ordered if:

(1) each $M_{n} (V)$ , $n \in N$ , is partially ordered;
(2) $γ^{*} M_{n} (V)^{+} γ \subseteq M_{m} (V)^{+}$ if $γ = [γ_{i j}]$ is any $n \times m$ matrix of complex numbers.

A matrix order unit space

(A, 1)

is a matrix ordered space

A

together with a distinguished order unit

1

satisfying the following conditions:

(1) $A^{+}$ is a proper cone with the order unit $1$ ;
(2) each of the cones $M_{n} (A)^{+}$ is Archimedean.

Each matrix order unit space

(A, 1)

may be provided with the norm

∥ a ∥ = inf {t \in R : [\begin{matrix} t 1 & a \\ a^{*} & t 1 \end{matrix}] \geq 0} .

In this paper, we will assume that

A

is complete for the norm. For a matrix order unit space

(A, 1)

, The matrix state space of

(A, 1)

is the collection

C S (A) = (C S_{n} (A))

of matrix states

C S_{n} (A) = {φ : φ

is a unital completely positive linear mapping from

A

into

M_{n}}

V

and

W

are vector spaces in duality, then they determine the matrix pairing

≪ \cdot, \cdot ≫ : M_{n} (V) \times M_{m} (W) \mapsto M_{n m},

where

≪ [v_{i j}], [w_{k l}] ≫ = [< v_{i j}, w_{k l} >]

for

[v_{i j}] \in M_{n} (V)

and

[w_{k l}] \in M_{m} (W)

A graded set

S = (S_{n})

is a sequence of sets

S_{n} (n \in N)

. If, for each

n \in N

M_{n} (V)

is a topological space, a graded set

S = (S_{n})

with

S_{n} \subseteq M_{n} (V)

is closed or compact if that is the case for each set

S_{n}

in the topology on

M_{n} (V)

. Given a vector space

V

, we say that a graded set

B = (B_{n})

with

B_{n} \subseteq M_{n} (V)

is absolutely matrix convex if for all

m, n \in N

B_{m} \oplus B_{n} \subseteq B_{m + n}

, and

α B_{m} β \subseteq B_{n}

for any contractions

α \in M_{n, m}

and

β \in M_{m, n}

. A matrix convex set in

V

is a graded set

K = (K_{n})

of subsets

K_{n} \subseteq M_{n} (V)

such that

\sum_{i = 1}^{k} γ_{i}^{*} v_{i} γ_{i} \in K_{n}

for all

v_{i} \in K_{n_{i}}

and

γ_{i} \in M_{n_{i}, n}

for

i = 1, 2, \dots, k

satisfying

\sum_{i = 1}^{k} γ_{i}^{*} γ_{i} = 1_{n}

. Let

V

and

W

be vector spaces in duality, and let

S = (S_{n})

be a graded set with

S_{n} \subseteq M_{n} (V)

. The absolute operator polar

S^{⊚} = (S_{n}^{⊚})

with

S_{n}^{⊚} \subseteq M_{n} (W)

, is defined by

S_{n}^{⊚} = {w \in M_{n} (W) : ∥ ≪ v, w ≫ ∥ \leq 1 f o r a l l v \in S_{r}, r \in N}

Given an arbitrary vector space

V

, a matrix gauge

G = (g_{n})

V

is a sequence of gauges

g_{n} : M_{n} (V) \mapsto [0, + \infty]

such that

(1) $g_{m + n} (v \oplus w) = max {g_{m} (v), g_{n} (w)}$ ;
(2) $g_{n} (α v β) \leq ∥ α ∥ g_{m} (v) ∥ β ∥$ ,

for any

v \in M_{m} (V)

w \in M_{n} (V)

α \in M_{n, m}

and

β \in M_{m, n}

. A matrix gauge

G = (g_{n})

is a matrix seminorm on

V

if for any

n \in N, g_{n} (v) < + \infty

for all

v \in M_{n} (V)

. If each

g_{n}

is a norm on

M_{n} (V)

, we say that

G

is a matrix norm. An operator space is a vector space together with a matrix norm on it. Given two operator spaces

V

and

W

. We denote by

C B (V, W)

the Banach space of all completely bounded linear mapping

φ

from

V

into

W

equipped with the completely bounded norm

∥ φ ∥_{c b}

. An operator system is a closed unital self-adjoint linear subspace of a unital

C^{*}

-algebra.

For a matrix order unit space

(A, 1)

, it is an operator space with the matrix norm determined by the matrix order on it. Every matrix order unit space is completely order isomorphic to an operator system[3] .

3 Continuous mappings on matrix metric spaces

First we recall the definition of matrix metrics on graded sets. See [16] for more details.

Definition 3.1. Let

V

be a vector space and let

K = (K_{n})

be a graded set with

K_{n} \subseteq M_{n} (V)

. A matrix metric

D = (D_{n})

K

is a sequence of metrics

D_{n} : K_{n} \times K_{n} \mapsto [0, + \infty)

such that

(1) if $x, u \in K_{m}$ and $y, v \in K_{n}$ such that $x \oplus y, u \oplus v \in K_{m + n}$ , then $D_{m + n} (x \oplus y, u \oplus v) = max {D_{m} (x, u), D_{n} (y, v)}$ ;
(2) if $x, u \in K_{m}$ and $α \in M_{m, n}$ with $α^{*} α = 1_{n}$ such that $α^{*} x α, α^{*} u α \in K_{n}$ , then $D_{n} (α^{*} x α, α^{*} u α) \leq D_{m} (x, u)$ .

The ordered pair

(K, D)

is said to be a matrix metric space over

V

Example 3.2. Let

V

be an operator space with matrix norm

∥ \cdot ∥ = (∥ \cdot ∥_{n})

and let

K = (K_{n})

be a graded set with

K_{n} \subseteq M_{n} (V)

. For

n \in N

and

x, y \in K_{n}

, we define

D_{n} (x, y) = ∥ x - y ∥_{n} .

Then

D = (D_{n})

is a matrix metric on

K

, and is called the matrix metric induced by the matrix norm on

V

(K, D)

is called a matrix metric space over the operator space

V

Example 3.3. Assume

ℒ = (L_{n})

is a matrix Lipschitz seminorm on the matrix order unit space

(V, 1)

(here we do not assume that

V

is complete for the matrix norm

∥ \cdot ∥ = (∥ \cdot ∥_{n})

determined by the matrix order on it) and the image of

L_{1}^{1} = {a \in V : L_{1} (a) \leq 1}

\tilde{V} = V / (C 1)

is totally bounded for

∥ \cdot ∥_{1}^{\sim}

. Then the sequence

D_{ℒ} = (D_{L_{n}})

of metrics defined by

D_{L_{n}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : L_{r} (a) \leq 1, r \in N},

for

φ, ψ \in C S_{n} (V)

and

n \in N

, is a matrix metric on

C S (V)

(see Theorem 5.3 in [15] ).

The theory of operator spaces is closely related to the structure of matrices over the spaces and the mappings on them. Here we extend the concept of mappings to this situation.

Definition 3.4. Let

V

and

W

be two vector spaces and let

K = (K_{n})

and

G = (G_{n})

be two graded sets with

K_{n} \subseteq M_{n} (V)

and

G_{n} \subseteq M_{n} (G)

. A matrix mapping from

K

into

G

is a sequence

f = (f_{n})

of mappings

f_{n} : K_{n} \mapsto G_{n}

. We denote by

ℳ (K, G)

the set of all matrix mappings from

K

into

G

When each

G_{n}

is a subspace of

M_{n} (G)

, we can define

f + g = (f_{n} + g_{n}), α f = (α f_{n}),

for

f = (f_{n}), g = (g_{n}) \in ℳ (K, G)

and

α \in C

. Then

ℳ (K, G)

is a vector space over

C

Now we define the matrix analogue of the equicontinuity.

Definition 3.5. Given two matrix metric spaces

(K_{1}, D_{1})

and

(K_{2}, D_{2})

, we say that a matrix mapping

f = (f_{n})

from

K_{1}

into

K_{2}

is continuous if each

f_{n}

is continuous. We let

C (K_{1}, K_{2})

denote the space of all continuous mappings

f : K_{1} \mapsto K_{2}

. A subset

S \subseteq C (K_{1}, K_{2})

is said to be equicontinuous if for any

ε > 0

n \in N

and

v \in K_{1, n}

there exists a

δ = δ (ε, v, n) > 0

such that

D_{2, n} (f_{n} (v), f_{n} (w)) < ε,

for all

f = (f_{k}) \in S

and

w \in K_{1, n}

with

D_{1, n} (v, w) < δ

. A subset

S \subseteq C (K_{1}, K_{2})

is said to be uniformly equicontinuous if for any

ε > 0

there exists a

δ = δ (ε) > 0

such that

D_{2, n} (f_{n} (v), f_{n} (w)) < ε,

for all

f = (f_{k}) \in S

and

n \in N

and

w, v \in K_{1, n}

with

D_{1, n} (v, w) < δ

A matrix mapping

f = (f_{n})

from matrix convex set

K = (K_{n})

into matrix convex set

G = (G_{n})

is said to be matrix affine if

f_{n} (\sum_{i = 1}^{k} γ_{i}^{*} v_{i} γ_{i}) = \sum_{i = 1}^{k} γ_{i}^{*} f_{n_{i}} (v_{i}) γ_{i}

for all

v_{i} \in K_{n_{i}}

and

γ_{i} \in M_{n_{i}, n}

for

i = 1, 2, \dots, k

satisfying

\sum_{i = 1}^{k} γ_{i}^{*} γ_{i} = 1_{n}

[14] . We let

A (K, G)

denote the set of all matrix affine mappings from

K

into

G

. Clearly the uniform equicontinuity of a subset

S \subseteq C (K_{1}, K_{2})

implies the equicontinuity of it. The following proposition indicates its converse holds in some special cases.

Proposition 3.6. Suppose that

(K, D)

is a matrix metric space and

K

is compact matrix convex. Let

(G, ℰ^{(k)})

be the matrix metric space with

G_{n} = M_{n} (M_{k})

and the matrix metric

ℰ^{(k)} = (E_{n}^{(k)})

induced by the matrix norm on

M_{k}

. Denote by

A C (K, G)

the set of all continuous and matrix affine mappings from

K

into

G

. If

S \subseteq A C (K, G)

is equicontinuous, the

S

is uniformly equicontinuous.

Proof. Given

ε > 0

. Since

S

is equicontinuous, for any

v \in K_{k}

we can find

δ = δ (ε, v) > 0

such that

E_{k}^{(k)} (f_{k} (v), f_{k} (w)) < \frac{ε}{8},

for all

f = (f_{n}) \in S

and

w \in K_{k}

with

D_{k} (v, w) < δ

. That

K_{k}

is compact means that there are

v_{1}, \dots, v_{t} \in K_{k}

such that

K_{k} \subseteq \cup_{i = 1}^{t} U (v_{i}; \frac{δ_{i}}{2})

, where

U (v_{i}; \frac{δ_{i}}{2}) = {w \in K_{k} : D_{k} (v_{i}, w) < \frac{δ_{i}}{2}}

and

δ_{i} = δ (ε, v_{i})

. Take

δ_{0} = \frac{1}{2} min {δ_{1}, \dots, δ_{t}}

. Then for

w, v \in K_{k}

with

D_{k} (v, w) < δ_{0}

, there is a

v_{i}

with

v \in U (v_{i}; \frac{δ_{i}}{2})

. So

D_{k} (w, v_{i}) \leq D_{k} (w, v) + D_{k} (v, v_{i}) < δ_{i}

. Hence for any

f = (f_{n}) \in S

we have

E_{k}^{(k)} (f_{k} (v), f_{k} (w)) \leq E_{k}^{(k)} (f_{k} (v), f_{k} (v_{i})) + E_{k}^{(k)} (f_{k} (v_{i}), f_{k} (w)) < \frac{ε}{4} .

Suppose that

v, w \in K_{r}

with

D_{r} (v, w) < δ_{0}

and

r \neq k

. If

r < k

, choose

u \in K_{k - r}

. Then

v \oplus u = [1_{r} 0_{r, k - r}]^{*} v [1_{r} 0_{r, k - r}] + [0_{k - r, r} 1_{k - r}]^{*} u [0_{k - r, r} 1_{k - r}] \in K_{k}

since

[1_{r} 0_{r, k - r}]^{*} [1_{r} 0_{r, k - r}] + [0_{k - r, r} 1_{k - r}]^{*} [0_{k - r, r} 1_{k - r}] = 1_{k}

and

K

is matrix convex. Similarly,

w \oplus u \in K_{k}

. Also

D_{k} (v \oplus u, w \oplus u) = D_{r} (v, w) < δ_{0}

. So for any

f = (f_{n}) \in S

we have

E_{k}^{(k)} (f_{k} (v \oplus u), f_{k} (w \oplus u)) < \frac{ε}{4} .

But that

f

is matrix affine implies

f_{k} (v \oplus u) = f_{r} (v) \oplus f_{k - r} (u), f_{k} (w \oplus u) = f_{r} (w) \oplus f_{k - r} (u),

that is,

E_{r}^{(k)} (f_{r} (v), f_{r} (w)) = E_{k}^{(k)} (f_{r} (v) \oplus f_{k - r} (u), f_{r} (w) \oplus f_{k - r} (u)) = E_{k}^{(k)} (f_{k} (v \oplus u), f_{k} (w \oplus u)) < \frac{ε}{4} < ε

. If

r > k

, for any unit vector

ξ \in C^{r} \otimes C^{k}

there exist an isometry

α : C^{k} \mapsto C^{r}

and a unit vector

ξ_{1} \in C^{k} \otimes C^{k}

such that

ξ = (α \otimes 1_{k}) (ξ_{1})

by Lemma 5.1 in [6] . Since

K

is matrix convex and

D_{k} (α^{*} v α, α^{*} w α) \leq D_{r} (v, w) < δ_{0}

, we have

\begin{matrix} | < (f_{r} (v) - f_{r} (w)) ξ, ξ > | \end{matrix}

\begin{matrix} = & | < (f_{r} (v) - f_{r} (w)) (α \otimes 1_{k}) (ξ_{1}), (α \otimes 1_{k}) (ξ_{1}) > | \end{matrix}

\begin{matrix} = & | < (α^{*} \otimes 1_{k}) (f_{r} (v) - f_{r} (w)) (α \otimes 1_{k}) (ξ_{1}), ξ_{1} > | \end{matrix}

\begin{matrix} = & | < (f_{k} (α^{*} v α) - f_{k} (α^{*} w α)) ξ_{1}, ξ_{1} > | \end{matrix}

\begin{matrix} \leq & ∥ f_{k} (α^{*} v α) - f_{k} (α^{*} w α) ∥ \end{matrix}

\begin{matrix} = & E_{k}^{(k)} (f_{k} (α^{*} v α), f_{k} (α^{*} w α)) \end{matrix}

\begin{matrix} < & \frac{ε}{4} . \end{matrix}

For any unit vectors

η, ζ \in C^{r} \otimes C^{k}

, we have

\begin{matrix} | < (f_{r} (v) - f_{r} (w)) η, ζ > | \end{matrix}

\begin{matrix} \leq & | < (f_{r} (v) - f_{r} (w)) \frac{η + ζ}{2}, \frac{η + ζ}{2} > | \end{matrix}

\begin{matrix} + | < (f_{r} (v) - f_{r} (w)) \frac{η - ζ}{2}, \frac{η - ζ}{2} > | \end{matrix}

\begin{matrix} + | i < (f_{r} (v) - f_{r} (w)) \frac{η + i ζ}{2}, \frac{η + i ζ}{2} > | \end{matrix}

\begin{matrix} + | i < (f_{r} (v) - f_{r} (w)) \frac{η - i ζ}{2}, \frac{η - i ζ}{2} > | \end{matrix}

\begin{matrix} < & ε . \end{matrix}

From the arbitrariness of

η

and

ζ

, we obtain that

E_{r}^{(k)} (f_{r} (v), f_{r} (w)) = ∥ f_{r} (v) - f_{r} (w) ∥ \leq ε

. By definition,

S

is uniformly equicontinuous. □

4 Matrix metric sets

Motivated by the idea of Antonescu and Christensen and our results in [16] , we give the operator space version of the metric set.

Definition 4.1. Let

(A, 1)

be a matrix order unit space. A graded set

K = (K_{n})

with

K_{n} \subseteq M_{n} (A)

is called a matrix metric set of

(A, 1)

if it is norm compact, self-adjoint and absolutely matix convex, and separates the matrix states on

(A, 1)

One can easily construct matrix metric sets for separable matrix order unit spaces.

Example 4.2. Given a countable group

G = {g_{n} : n \in N}

and a closed self-adjoint subspace

A

C_{r}^{*} (G)

containing the unit

λ_{e}

, where

e

is the identity element of

G

. Then with the usual partial ordering on

M_{n} (A)

for

n \in N

(A, λ_{e})

becomes a matrix order unit space. Set

K = A \cap \bar{c o} (\cup_{n = 1}^{\infty} {α λ_{g_{n}} + β λ_{g_{n}}^{*} : | α | + | β | \leq \frac{1}{n}, α, β \in C}),

where

\bar{c o}

means the closed convex hull. Then

K

is a norm compact, self-adjoint and absolutely convex subset of

A

, and so is weakly closed and absolutely convex. Thus there is a weakly closed absolutely matrix convex set

K = (K_{n})

with

K_{n} \subseteq M_{n} (A)

and

K_{1} = K

(see page 181 in [6] ).

Clearly

K

is norm closed. For any

n \in N

and

a = [a_{i j}] \in K_{n}

, we have

a_{i j} = [0 \dots 01_{i} 0 \dots 0] a [0 \dots 01_{j} 0 \dots 0]^{*} \in K_{1} .

Since

K_{1} = K

is norm compact,

K_{1}

is totally bounded, and hence every

K_{n}

is totally bounded. Each

K_{n}

is norm closed implies it is also norm compact. Clearly

K

separates the states on

A

. Since the matrix state space

C S (A)

A

is matrix convex,

K

also separates the matrix states on

A

. So

K

separates the matrix states on

A

. Therefore,

K

is a matrix metric set of

(A, λ_{e})

The natural topology on the matrix state space

C S (A)

is the BW-topology, that is, topologies each

C S_{n} (A)

by BW-topology (see page 146 in [2] ). The following result justifies the definition of a matrix metric set, that is, it generates the BW-topology.

Proposition 4.3. Let

(A, 1)

be a matrix order unit space,

C S (A)

the matrix state space of

(A, 1)

and

K = (K_{n})

a matrix metric set of

(A, 1)

Then

D_{K} = (D_{K_{n}})

, where

D_{K_{n}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : a \in K_{r}, r \in N},

for

φ, ψ \in C S_{n} (A)

and

n \in N

, is a matrix metric on

C S (A)

and the

D_{K}

topology on

C S (A)

agrees with the BW-topology.

Proof. Denote

C_{n} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : a \in K_{n}},

for

φ, ψ \in C S_{n} (A)

and

n \in N

. Clearly

C_{n} (φ, ψ) \leq D_{K_{n}} (φ, ψ)

for all

φ, ψ \in C S_{n} (A)

and

n \in N

For

a = [a_{p q}] \in K_{r}

and

r < n

, we have that

0_{n - r} = 0_{n - r, r} a 0_{r, n - r} \in K_{n - r}

and hence

a \oplus 0_{n - r} \in K_{n}

because

K

is absolutely matrix convex. So for

φ, ψ \in C S_{n} (A)

, we have

\begin{matrix} ∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ \end{matrix}

\begin{matrix} = & ∥ ≪ φ, a \oplus 0_{n - r} ≫ - ≪ ψ, a \oplus 0_{n - r} ≫ ∥ \leq C_{n} (φ, ψ) . \end{matrix}

Suppose that

r > n

. For arbitrary unit vectors

ξ, η \in C^{r} \otimes C^{n}

, there exist isometries

α, β : C^{n} \mapsto C^{r}

and unit vectors

ξ_{1}, η_{1} \in C^{n} \otimes C^{n}

for which

ξ = (α \otimes 1_{n}) (ξ_{1})

and

η = (β \otimes 1_{n}) (η_{1})

by Lemma 5.1 in [6] . That

K

is absolutely matrix convex implies

\begin{matrix} | < (≪ φ, a ≫ - ≪ ψ, a ≫) η, ξ > | \end{matrix}

\begin{matrix} = & | < (≪ φ, a ≫ - ≪ ψ, a ≫) (β \otimes 1_{n}) (η_{1}), (α \otimes 1_{n}) (ξ_{1}) > | \end{matrix}

\begin{matrix} = & | < (≪ φ, α^{*} a β ≫ - ≪ ψ, α^{*} a β ≫) η_{1}, ξ_{1} > | \end{matrix}

\begin{matrix} \leq & ∥ ≪ φ, α^{*} a β ≫ - ≪ ψ, α^{*} a β ≫ ∥ \end{matrix}

\begin{matrix} \leq & C_{n} (φ, ψ) . \end{matrix}

Since

ξ

and

η

are arbitrary unit vectors, we conclude that

∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ \leq C_{n} (φ, ψ)

. Therefore,

D_{K_{n}} (φ, ψ) \leq C_{n} (φ, ψ)

for all

φ, ψ \in C S_{n} (A)

and

n \in N

, and so

C_{n} (φ, ψ) = D_{K_{n}} (φ, ψ)

for all

φ, ψ \in C S_{n} (A)

and

n \in N

. Since

K

separates

C S (A)

and norm compact, each

D_{K_{n}}

is a bounded metric on

C S_{n} (A)

For

φ_{1}, φ_{2} \in C S_{m} (A)

and

ψ_{1}, ψ_{2} \in C S_{p} (A)

, we have that

φ_{1} \oplus ψ_{1}, φ_{2} \oplus ψ_{2} \in C S_{m + p} (A)

and

\begin{matrix} D_{K_{m + p}} (φ_{1} \oplus ψ_{1}, φ_{2} \oplus ψ_{2}) \end{matrix}

\begin{matrix} = & sup {∥ ≪ φ_{1} \oplus ψ_{1}, a ≫ - ≪ φ_{2} \oplus ψ_{2}, a ≫ ∥ : a \in K_{r}, r \in N} \end{matrix}

\begin{matrix} = & sup {max {∥ ≪ φ_{1}, a ≫ - ≪ φ_{2}, a ≫ ∥, \end{matrix}

\begin{matrix} ∥ ≪ ψ_{1}, a ≫ - ≪ ψ_{2}, a ≫ ∥} : a \in K_{r}, r \in N} \end{matrix}

\begin{matrix} = & max {sup {∥ ≪ φ_{1}, a ≫ - ≪ φ_{2}, a ≫ ∥, a \in K_{r}, r \in N}, \end{matrix}

\begin{matrix} sup {∥ ≪ ψ_{1}, a ≫ - ≪ ψ_{2}, a ≫ ∥ : a \in K_{r}, r \in N}} \end{matrix}

\begin{matrix} = & max {D_{K_{m}} (φ_{1}, φ_{2}), D_{K_{p}} (ψ_{1}, ψ_{2})} . \end{matrix}

φ, ψ \in C S_{m} (A)

and

α \in M_{m, p}

with

α^{*} α = 1_{n}

, then

α^{*} φ α, α^{*} ψ α \in C S_{p} (A)

, and

\begin{matrix} D_{K_{p}} (α^{*} φ α, α^{*} ψ α) \end{matrix}

\begin{matrix} = & sup {∥ ≪ α^{*} φ α, a ≫ - ≪ α^{*} ψ α, a ≫ ∥ : a \in K_{r}, r \in N} \end{matrix}

\begin{matrix} = & sup {∥ (α^{*} \otimes 1_{r}) (≪ φ, a ≫ - ≪ ψ, a ≫) (α \otimes 1_{r}) ∥ : a \in K_{r}, r \in N} \end{matrix}

\begin{matrix} \leq & sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥, a \in K_{r}, r \in N} \end{matrix}

\begin{matrix} = & D_{K_{m}} (φ, ψ)} . \end{matrix}

Therefore,

D_{K}

is a matrix metric on

C S (A)

Clearly, the topology on each

C S_{n} (A)

induced by

D_{K_{n}}

is a Hausdorff topology.

Suppose

{φ_{i}} \subseteq C S_{n} (A)

φ \in C S_{n} (A)

and

{lim}_{i} φ_{i} = φ

in the BW-topology. Then

{lim}_{i} φ_{i} (a) = φ (a)

for all

a \in A

. Given

ε > 0

. For

a \in K_{1}

, there is an

I_{a}

such that

∥ φ_{i} (a) - φ (a) ∥ < \frac{ε}{3 n^{2}}

for

i \geq I_{a}

. When

b \in U (a; \frac{ε}{3 n^{2}}) = {c \in A : ∥ c - a ∥ < \frac{ε}{3 n^{2}}}

, we have

\begin{matrix} ∥ φ_{i} (b) - φ (b) ∥ \end{matrix}

\begin{matrix} \leq & ∥ φ_{i} (b) - φ_{i} (a) ∥ + ∥ φ_{i} (a) - φ (a) ∥ + ∥ φ (a) - φ (b) ∥ \end{matrix}

\begin{matrix} \leq & 2 ∥ b - a ∥ + ∥ φ_{i} (a) - φ (a) ∥ \end{matrix}

\begin{matrix} < & \frac{ε}{n^{2}}, \end{matrix}

for

i \geq I_{a}

. Since

K_{1}

is norm compact, there exists an

s \in N

such that

K_{1} \subseteq \cup_{i = 1}^{s} U (a_{i}; \frac{ε}{3 n^{2}})

for some

a_{1}, \dots, a_{s} \in K_{1}

. Then for

i \geq I_{a_{j}}, j = 1, 2, \dots, s

, and

c \in K_{1}

, we can find an

i_{0} \in {1, 2, \dots, s}

such that

c \in U (a_{i_{0}}, \frac{ε}{3 n^{2}})

, and so

∥ φ_{i} (c) - φ (c) ∥ < \frac{ε}{n^{2}}

. For

a = [a_{p q}] \in K_{n}

and

i \geq I_{a_{j}}, j = 1, 2, \dots, s

, we obtain

\begin{matrix} ∥ ≪ φ_{i}, a ≫ - ≪ φ, a ≫ ∥ & = & ∥ [φ_{i} (a_{p q}) - φ (a_{p q})] ∥ \end{matrix}

\begin{matrix} \leq & \sum_{p, q = 1}^{n} ∥ φ_{i} (a_{p q}) - φ (a_{p q}) ∥ < ε, \end{matrix}

Therefore,

D_{K_{n}} (φ_{i}, ψ) \leq ε

for

i \geq I_{a_{j}}, j = 1, 2, \dots, s

, that is,

{lim}_{i} φ_{i} = φ

in the

D_{K_{n}}

-topology. So

D_{K}

-topology on

C S (A)

is weaker than the BW-topology. On the other hand,

C S (A)

is BW-compact by Theorem 6.4 in [8] , and so

D_{K}

-topology and BW-topology agree. □

From the proof of Proposition 4.3 , we have

Corollary 4.4. Let

(A, 1)

be a matrix order unit space,

C S (A)

the matrix state space of

(A, 1)

and

K = (K_{n})

a matrix metric set of

(A, 1)

Then

D_{K_{n}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : a \in K_{n}},

for

φ, ψ \in C S_{n} (A)

and

n \in N

By a matrix Lip-gauge on a matrix order unit space

(A, 1)

we mean a matrix gauge

G = (G_{n})

(A, 1)

such that: (1) the null space of each

G_{n}

M_{n} (C 1)

; (2)

G_{n} (v^{*}) = G_{n} (v)

for any

v \in M_{n} (A)

; (3)

{v \in A : G_{1} (v) < + \infty}

is dense in

A

; (4) the

D_{G}

-topology on

C S (A)

agrees with the BW-topology. The matrix Lip-gauge

G = (G_{n})

is lower semicontinuous if each

G_{n}

is lower semicontinuous.

Corollary 4.5. Let

(A, 1)

be a matrix order unit space,

C S (A)

the matrix state space of

(A, 1)

and

K = (K_{n})

a matrix metric set of

(A, 1)

Then there is a lower semicontinuous matrix Lip-gauge

ℒ = (L_{n})

(A, 1)

such that

D_{K_{n}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : L_{r} (a) \leq 1, r \in N},

for

φ, ψ \in C S_{n} (A)

and

n \in N

Proof. Clearly $D_{K}$ is convex, midpoint balance, and midpoint concave. Now the corollary follows from Theorem 6.12 in [16] . □

5 Relative compactness

In this section we state and prove our main results. First let us take a look at what is the boundedness of matrix mappings.

Definition 5.1. Given two matrix metric spaces

(K_{1}, D_{1})

and

(K_{2}, D_{2})

Fix an

x = (x_{n}) \in K_{2}

(that is, each

x_{n} \in K_{2, n}

). For a matrix mapping

f = (f_{n}) \in ℳ (K_{1}, K_{2})

, we define

p (f) = sup {D_{2, n} (f_{n} (w), x_{n}) : w \in K_{1, n}, n \in N} .

p (f) < + \infty

, we say that

f

is bounded. Given

S \subseteq ℳ (K_{1}, K_{2})

. If there is a constant

C > 0

such that

p (f) \leq C

for any

f \in S

, we say that

S

is bounded.

(K_{2}, D_{2})

is the matrix metric space over an operator space, we take

x = (0_{n})

Then clearly

p (f) = sup {∥ f_{n} (w) ∥_{n} : w \in K_{1, n}, n \in N}

is a faithful gauge on

ℳ (K_{1}, K_{2})

. The following lemma displays one of their aspects of the boundedness of matrix mappings.

Lemma 5.2. Let

(A, 1)

be a matrix order unit space. The canonical mapping of

A

into

A (C S (A))

, which sends

x \in M_{n} (A)

\hat{x} = ({\hat{x}}_{r}) \in M_{n} (A (C S (A))) ≃ A (C S (A), M_{n})

given by

{\hat{x}}_{r} (φ) = ≪ φ, x ≫

for

φ \in C S_{r} (A)

and

r \in N

, is a unital matrix order preserving bijection between

A

and

A (C S (A))

(

see page

314

in [14]

)

. Define

p_{n} (\hat{x}) = sup {∥ ≪ φ, x ≫ ∥ : φ \in C S_{r} (A), r \in N},

for

x \in M_{n} (A)

and

n \in N

. Then each

p_{n}

is a norm on

A (C S (A), M_{n})

Moreover,

p_{n} (\hat{x}) = ∥ x ∥_{n},

for

x \in M_{n} (A)

and

n \in N

Proof. Since

p_{n} (\hat{x}) = sup {∥ {\hat{x}}_{r} (φ) ∥ : φ \in C S_{r} (A), r \in N},

p_{n}

is a faithful gauge. For

x \in M_{n} (A)

, we have

\begin{matrix} ∥ x ∥_{n} & = & inf {t \in R : [\begin{matrix} t 1_{n} & x \\ x^{*} & t 1_{n} \end{matrix}] \geq 0} \end{matrix}

\begin{matrix} = & inf {t \in R : ≪ φ, [\begin{matrix} t 1_{n} & x \\ x^{*} & t 1_{n} \end{matrix}] ≫ \geq 0, φ \in C S_{r} (A), r \in N} \end{matrix}

\begin{matrix} = & inf {t \in R : [\begin{matrix} t 1_{r} \otimes 1_{n} & ≪ φ, x ≫ \\ ≪ φ, x ≫^{*} & t 1_{r} \otimes 1_{n} \end{matrix}] \geq 0, φ \in C S_{r} (A), r \in N} \end{matrix}

\begin{matrix} = & sup {∥ ≪ φ, x ≫ ∥ : φ \in C S_{r} (A), r \in N} \end{matrix}

\begin{matrix} = & p_{n} (\hat{x}) . \end{matrix}

□

Lemma 5.3. With notation as in Lemma 5.2

, we have

\frac{1}{4} p_{n} (\hat{x}) \leq q_{n} (\hat{x}) \leq p_{n} (\hat{x}),

for

x \in M_{n} (A)

and

n \in N

, where

q_{n} (\hat{x}) = sup {∥ ≪ φ, x ≫ ∥ : φ \in C S_{n} (A)}

Proof. Clearly,

q_{n} (\hat{x}) \leq p_{n} (\hat{x})

. Given

φ \in C S_{r} (A)

. If

r < n

, we have

\begin{matrix} ψ & = & [1_{r} 0_{r, n - r}]^{*} φ [1_{r} 0_{r, n - r}] \end{matrix}

\begin{matrix} + [0_{n - r, r} 1_{n - r}]^{*} ([10_{1, r - 1}] φ [10_{1, r - 1}]^{*} \otimes 1_{n - r}) [0_{n - r, r} 1_{n - r}] \in C S_{n} (A), \end{matrix}

and so

q_{n} (\hat{x}) \geq ∥ ≪ ψ, x ≫ ∥ \geq ∥ ≪ φ, x ≫ ∥ .

r > n

, for any unit vector

ξ \in C^{r} \otimes C^{n}

there exist an isometry

α : C^{n} \mapsto C^{r}

and a unit vector

ξ_{1} \in C^{n} \otimes C^{n}

for which

(α \otimes 1_{n}) (ξ_{1}) = ξ

. So we get

\begin{matrix} | < ≪ φ, x ≫ ξ, ξ > | & = & | < ≪ φ, x ≫ (α \otimes 1_{n}) (ξ_{1}), (α \otimes 1_{n}) (ξ_{1}) > | \end{matrix}

\begin{matrix} = & | < ≪ α^{*} φ α, x ≫ ξ_{1}, ξ_{1} > | \end{matrix}

\begin{matrix} \leq & ∥ ≪ α^{*} φ α, x ≫ ∥ \leq q_{n} (\hat{x}) . \end{matrix}

Now for any unit vectors

η, ζ \in C^{r} \otimes C^{n}

, we have

\begin{matrix} | < ≪ φ, x ≫ η, ζ > | \end{matrix}

\begin{matrix} \leq & | < ≪ φ, x ≫ \frac{η + ζ}{2}, \frac{η + ζ}{2} > | + | < ≪ φ, x ≫ \frac{η - ζ}{2}, \frac{η - ζ}{2} > | \end{matrix}

\begin{matrix} + | i < ≪ φ, x ≫ \frac{η + i ζ}{2}, \frac{η + i ζ}{2} > | + | i < ≪ φ, x ≫ \frac{η - i ζ}{2}, \frac{η - i ζ}{2} > | \end{matrix}

\begin{matrix} \leq & 4 q_{n} (\hat{x}) . \end{matrix}

Thus

∥ ≪ φ, x ≫ ∥ \leq 4 q_{n} (\hat{x})

. Since

φ

is arbitrary, we obtain

\frac{1}{4} p_{n} (\hat{x}) \leq q_{n} (\hat{x})

. □

For a matrix order unit space

(A, 1)

and

ε > 0

, we denote

C 1 = (M_{n} (C 1))

and

A_{ε} = (A_{n, ε})

with

A_{n, ε} = {a \in M_{n} (A) : ∥ a ∥_{n} \leq ε}

for

n \in N

, where

∥ \cdot ∥ = (∥ \cdot ∥_{n})

is the matrix norm determined by the matrix order on

(A, 1)

. Given two graded sets

U = (U_{n})

and

V = (V_{n})

with

U_{n}, V_{n} \in M_{n} (A)

, and a sequence

λ = (λ_{n})

with

λ_{n} \in C

, we define

U + V = (U_{n} + V_{n}), λ U = (λ_{n} U_{n}) .

If all

λ_{n} = λ

, we define

λ U = (λ U_{n})

. Let

(K, D)

be a matrix metric space and

G

a graded set with

G \subseteq K

, that is, each

G_{n} \subseteq K_{n}

. If every

G_{n}

is bounded with respect to the metric

D_{n}

, we say that

G

is bounded.

Theorem 5.4. Let

(A, 1)

be a matrix order unit space and

K

a matrix metric set of

(A, 1)

. For any graded set

S = (S_{n})

with

S_{n} \subseteq M_{n} (A)

, the following conditions are equivalent:

(1) The graded set $S$ is norm relatively compact.
(2) The graded set $\hat{S} = ({\hat{S}}_{n})$ , where ${\hat{S}}_{n} = {\hat{a} \in A (C S (A), M_{n}) : a \in S_{n}}$ , is bounded and equicontinuous with respect to the matrix metric $D_{K}$ and the matrix metric $ℰ^{(n)}$ induced by the matrix norm on each $M_{n}$ .
(3) The graded set $S$ is bounded and for every $ε > 0$ there exists a sequence $λ = (λ_{n})$ with $λ_{n} > 0$ such that $S \subseteq A_{ε} + λ K + C 1 .$

Proof. We will show

(1) ⟹ (2) ⟹ (3) ⟹ (1)

. Assume (1). Then each

S_{n}

is norm relatively compact. Since

M_{n} (A)

is complete, each

S_{n}

is totally bounded. By Lemma 5.2 , each

{\hat{S}}_{n}

is totally bounded, and so each

{\hat{S}}_{n}

is bounded.

Given

a = [a_{i j}] \in S_{n}

and

ε > 0

. For any

k \in N

, suppose that

φ \in C S_{k} (A), {φ_{s}} \subseteq C S_{k} (A)

and

{lim}_{s} φ_{s} = φ

in the

D_{K_{k}}

-topology. By Proposition 4.3 ,

D_{K_{k}}

-topology on

C S_{k} (A)

agrees with the BW-topology. So there exists an

s_{0}

such that when

s > s_{0}

, we have

∥ φ_{s} (a_{i j}) - φ (a_{i j}) ∥ < \frac{ε}{n^{2}},

for

i, j = 1, 2, \dots, n

. Now for

s > s_{0}

, we have

\begin{matrix} E_{k}^{(n)} ({\hat{a}}_{k} (φ_{s}), {\hat{a}}_{k} (φ)) & = & ∥ {\hat{a}}_{k} (φ_{s}) - {\hat{a}}_{k} (φ) ∥ \end{matrix}

\begin{matrix} = & ∥ ≪ φ_{s}, a ≫ - ≪ φ, a ≫ ∥ \end{matrix}

\begin{matrix} = & ∥ [φ_{s} (a_{i j}) - φ (a_{i j})] ∥ \end{matrix}

\begin{matrix} \leq & \sum_{i, j = 1}^{n} ∥ φ_{s} (a_{i j}) - φ (a_{i j}) ∥ < ε . \end{matrix}

Hence

\hat{a} \in C (C S (A), M_{n})

Fix

ε > 0, k \in N

and

φ \in C S_{k} (A)

. Since

{\hat{S}}_{n}

is totally bounded, there is an

\frac{ε}{3}

-net

{{\hat{a}}^{(1)}, \dots, {\hat{a}}^{(t)}}

{\hat{S}}_{n}

{{\hat{a}}^{(1)}, \dots, {\hat{a}}^{(t)}} \subseteq C (C S (A), M_{n})

implies that there is a

δ = δ (ε, φ, k) > 0

such that

E_{k}^{(n)} ({\hat{a}}_{k}^{(s)} (φ), {\hat{a}}_{k}^{(s)} (ψ)) < \frac{ε}{3}, s = 1, 2, \dots, t,

for

ψ \in C S_{k} (A)

with

D_{K_{k}} (φ, ψ) < δ

. For any

\hat{a} \in {\hat{S}}_{n}

, there is a

{\hat{a}}^{(s)} \in {{\hat{a}}^{(1)}, \dots, {\hat{a}}^{(t)}}

such that

p_{n} (\hat{a} - {\hat{a}}^{(s)}) < \frac{ε}{3}

. So for

ψ \in C S_{k} (A)

with

D_{K_{k}} (φ, ψ) < δ

, we have

\begin{matrix} E_{k}^{(n)} ({\hat{a}}_{k} (φ), {\hat{a}}_{k} (ψ)) & = & ∥ {\hat{a}}_{k} (φ) - {\hat{a}}_{k} (ψ) ∥ \end{matrix}

\begin{matrix} \leq & ‖ {\hat{a}}_{k} (φ) - {\hat{a}}_{k}^{(s)} (φ) ‖ + ‖ {\hat{a}}_{k}^{(s)} (φ) - {\hat{a}}_{k}^{(s)} (ψ) ‖ \end{matrix}

\begin{matrix} + ‖ {\hat{a}}_{k}^{(s)} (ψ) - {\hat{a}}_{k} (ψ) ‖ \end{matrix}

\begin{matrix} \leq & 2 {‖ a - a^{(s)} ‖}_{n} + E_{k}^{(n)} ({\hat{a}}_{k}^{(s)} (φ), {\hat{a}}_{k}^{(s)} (ψ)) \end{matrix}

\begin{matrix} = & 2 p_{n} (\hat{a} - {\hat{a}}^{(s)}) + E_{k}^{(n)} ({\hat{a}}_{k}^{(s)} (φ), {\hat{a}}_{k}^{(s)} (ψ)) \end{matrix}

\begin{matrix} < & ε . \end{matrix}

By definition,

{\hat{S}}_{n}

is equicontinuous, whence (2).

Assume (2). From the boundedness of

\hat{S}

it follows that

S

is bounded by Lemma 5.2 . Since the

D_{K}

-topology on

C S (A)

agrees with the BW-topology (see Proposition 4.3 ),

C S (A)

is compact in the

D_{K}

-topology. Moreover,

C S (A)

is matrix convex.

Clearly

{\hat{S}}_{n} \subseteq A C (C S (A), M_{n})

. By Proposition 3.6 , each

{\hat{S}}_{n}

is uniformly equicontinuous because each

{\hat{S}}_{n}

is equicontinuous. Given

ε > 0

and

n \in N

. We can find a

δ_{n} > 0

such that

E_{k}^{(n)} ({\hat{a}}_{k} (φ), {\hat{a}}_{k} (ψ)) < ε,

for

a \in S_{n}

and

φ, ψ \in C S_{k} (A)

with

D_{K_{k}} (φ, ψ) \leq δ_{n}

and

k \in N

. Because

φ - ψ \in K_{k}^{⊚}

if and only if

D_{K_{k}} (φ, ψ) \leq 1

D_{K_{k}} (φ, ψ) \leq δ_{n}

if and only if

φ - ψ \in δ_{n} (K_{k}^{⊚})

So we have

∥ ≪ φ - ψ, a ≫ ∥ < ε,

for

a \in S_{n}

and

φ, ψ \in C S_{k} (A)

with

φ - ψ \in δ_{n} (K_{k}^{⊚})

and

k \in N

For

f \in M_{k} (A^{*}) ≃ C B (A, M_{k})

with

∥ f ∥_{c b} \leq 2

and

≪ f, c ≫ = 0_{r k}

for

c \in M_{r} (C 1)

and

r \in N

, there are

φ_{1}, φ_{2}, φ_{3}, φ_{4} \in C S_{k} (A)

such that

f = φ_{1} - φ_{2} + i (φ_{3} - φ_{4})

by Lemma 4.1 in [16] . If

f \in δ_{n} (K_{k}^{⊚})

, then

\begin{matrix} ∥ ≪ φ_{1} - φ_{2}, a ≫ ∥ & = & ‖ ≪ \frac{1}{2} (f + f^{*}), a ≫ ‖ \end{matrix}

\begin{matrix} \leq & \frac{1}{2} (∥ ≪ f, a ≫ ∥ + ∥ ≪ f^{*}, a ≫ ∥) \end{matrix}

\begin{matrix} = & \frac{1}{2} (∥ ≪ f, a ≫ ∥ + ∥ ≪ f, a^{*} ≫^{*} ∥) \end{matrix}

\begin{matrix} \leq & δ_{n}, \end{matrix}

for

a \in K_{r}

and

r \in N

. We obtain that

φ_{1} - φ_{2} \in δ_{n} (K_{k}^{⊚})

. Similarly, we have

φ_{3} - φ_{4} \in δ_{n} (K_{k}^{⊚})

. So for

a \in S_{n}

we have

∥ ≪ f, a ≫ ∥ < 2 ε .

Denote

{C 1}_{k}^{⊥} = {f \in M_{k} (A^{*}) : ≪ f, c ≫ = 0_{k r}

for

c \in M_{r} (C 1), r \in N}

and

T_{k} = (A^{*})_{k, 2} \cap {C 1}_{k}^{⊥} \cap δ_{n} (K_{k}^{⊚})

for

k \in N

. Let

T = (T_{k})

. We have

\begin{matrix} T_{n}^{⊚} & = & {b \in M_{n} (A) : ∥ ≪ f, b ≫ ∥ \leq 1, f \in T_{k}, k \in N} \end{matrix}

\begin{matrix} = & \frac{1}{2 ε} {b \in M_{n} (A) : ∥ ≪ f, b ≫ ∥ \leq 2 ε, f \in T_{k}, k \in N} . \end{matrix}

Thus if

a \in S_{n}

, then

a \in 2 ε T_{n}^{⊚}

. So

S_{n} \subseteq 2 ε T_{n}^{⊚}

. Set

W_{k} = A_{k, \frac{1}{2}} \cup M_{k} (C 1) \cup \frac{1}{δ_{n}} K_{k}, k \in N,

and

W = (W_{k})

. We have

W_{n}^{⊚} = {f \in M_{n} (A^{*}) : ∥ ≪ f, b ≫ ∥ \leq 1, b \in W_{k}, k \in N} = T_{n} .

S_{n} \subseteq 2 ε W_{n}^{⊚ ⊚}

. Since

A_{k, \frac{1}{2}}

and

M_{k} (C 1)

are norm closed and

\frac{1}{δ_{n}} K_{k}

is norm compact,

X_{k} = A_{k, \frac{1}{2}} + M_{k} (C 1) + \frac{1}{δ_{n}} K_{k}

is norm closed. Clearly

X = (X_{k})

is absolutely matrix convex. So

X

is weakly closed,

W \subseteq X

and

X

is contained in any absolutely matrix convex set containing

W

. But the generalized bipolar theorem says that

W^{⊚ ⊚}

equals

\bar{a m c o} (W)

, the smallest weakly closed absolutely matrix convex set containing

W

(see Proposition 4.1 in [6] ). Therefore,

S_{n} \subseteq 2 ε W_{n}^{⊚ ⊚} = 2 ε X_{n} = A_{n, ε} + M_{n} (C 1) + λ_{n} K_{n}

, where

λ_{n} = \frac{2 ε}{δ_{n}}

, whence (3).

Assume (3). Then

S

is bounded, and so

\hat{S}

is bounded by Lemma 5.2 . Given

ε > 0

. We can find a sequence

λ = (λ_{n})

with

λ_{n} > 0

such that

S \subseteq A_{\frac{ε}{128}} + λ K + C 1 .

For

n \in N

and

a \in S_{n}

, there are

b \in A_{n, \frac{ε}{128}}, c \in K_{n}

and

α = [α_{i j}] \in M_{n}

such that

a = b + λ_{n} c + [α_{i j} 1]

. Then for

k \in N

and

φ, ψ \in C S_{k} (A)

with

D_{K_{k}} (φ, ψ) < \frac{ε}{64 λ_{n}}

, we have

\begin{matrix} E_{k}^{(n)} ({\hat{a}}_{k} (φ), {\hat{a}}_{k} (ψ)) \end{matrix}

\begin{matrix} = & ∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ \end{matrix}

\begin{matrix} \leq & ∥ ≪ φ - ψ, b ≫ ∥ + | λ_{n} | ∥ ≪ φ, c ≫ - ≪ ψ, c ≫ ∥ \end{matrix}

\begin{matrix} \leq & 2 ∥ b ∥ + | λ_{n} | D_{K_{k}} (φ, ψ) < \frac{ε}{32} . \end{matrix}

C S_{n} (A)

is BW-compact and

C S_{n} (A) = \cup_{φ \in C S_{n} (A)} U (φ; \frac{ε}{64 λ_{n}})

. Hence there are

φ_{1}, \dots, φ_{m} \in C S_{n} (A)

such that

C S_{n} (A) = \cup_{i = 1}^{m} U (φ_{i}; \frac{ε}{64 λ_{n}})

. Since

\hat{S}

is bounded, there exists an

M_{n} > 0, n \in N

, such that

∥ {\hat{a}}_{n} (φ) ∥ \leq M_{n}

for all

φ \in C S_{n} (A)

and

a \in S_{n}

. So

{{\hat{a}}_{n} (φ) : φ \in C S_{n} (A), a \in S_{n}}

is totally bounded. The sets

V_{i} = {{\hat{a}}_{n} (φ_{i}) : a \in S_{n}}, i = 1, 2, \dots, m

, are all totally bounded. Thus there are

\frac{ε}{32}

-nets in each of them which are denoted by some sets

{x_{i 1}, \dots, x_{i k_{i}}}

where

i = 1, 2, \dots, m

. Hence we obtain

\cup_{i = 1}^{m} V_{i} \subseteq \cup_{i = 1}^{m} \cup_{l_{i} = 1}^{k_{i}} U (x_{i l_{i}}; \frac{ε}{32}) = \cup_{j = 1}^{N} U (z_{j}; \frac{ε}{32}),

where

N = k_{1} + k_{2} + \dots + k_{m}

and the sets

{1, \dots, k_{1}}

{k_{1} + 1, \dots, k_{1} + k_{2}}

\dots

{k_{1} + \dots + k_{m - 1} + 1, \dots, N}

of the index

j

correspond, respectively, to the sets

{11, \dots, 1 k_{1}}

{21, \dots, 2 k_{2}}

\dots

{m 1, \dots, m k_{m}}

of the pair of indices

i

and

l_{i}

. The points

z_{j}

are none other than the points

x_{i l_{i}}

relabeled in this way.

Let

Γ

denote the finite set of all mappings from

{1, 2, \dots, m}

into

{1, 2, \dots, N}

For any

γ \in Γ

, we denote

O_{γ} = {a \in S_{n} : E_{n}^{(n)} ({\hat{a}}_{n} (φ_{1}), z_{γ (1)}) < \frac{ε}{32}, \dots, E_{n}^{(n)} ({\hat{a}}_{n} (φ_{m}), z_{γ (m)}) < \frac{ε}{32}} .

Evidently, we have

S_{n} = \cup_{γ \in Γ} O_{γ}

. For

γ \in Γ

and

a, b \in O_{γ}

and

φ \in C S_{n} (A)

, there exists a

φ_{i}

such that

D_{K_{n}} (φ, φ_{i}) < \frac{ε}{64 λ_{n}}

. We obtain

\begin{matrix} ∥ ≪ φ, a - b ≫ ∥ \end{matrix}

\begin{matrix} = & E_{n}^{(n)} ({\hat{a}}_{n} (φ), {\hat{b}}_{n} (φ)) \end{matrix}

\begin{matrix} \leq & E_{n}^{(n)} ({\hat{a}}_{n} (φ), {\hat{a}}_{n} (φ_{i})) + E_{n}^{(n)} ({\hat{a}}_{n} (φ_{i}), z_{γ (i)}) \end{matrix}

\begin{matrix} + E_{n}^{(n)} (z_{γ (i)}, {\hat{b}}_{n} (φ_{i})) + E_{n}^{(n)} ({\hat{b}}_{n} (φ_{i}), {\hat{b}}_{n} (φ)) \end{matrix}

\begin{matrix} < & \frac{ε}{8} . \end{matrix}

By the arbitrariness of

φ

, we have

q_{n} (\hat{a - b}) = sup {∥ ≪ φ, a - b ≫ ∥ : φ \in C S_{n} (A)} \leq \frac{ε}{8} .

In view of Lemma 5.3 , we get

p_{n} (\hat{a - b}) \leq \frac{ε}{2}

. So

∥ a - b ∥_{n} \leq \frac{ε}{2}

by Lemma 5.2 . Hence for any

γ \in Γ

there exists an

a_{γ} \in S_{n}

such that

O_{γ} \subseteq U (a_{γ}; ε)

. So

S_{n} \subseteq \cup_{γ \in Γ} U (a_{γ}; ε)

, namely

S_{n}

is totally bounded. Since

M_{n} (A)

is complete,

S_{n}

is norm relatively compact, whence (1). □

Now we come to the “function” level. In [1] , the point of view on Lip-norms is also nearly the same as the one Kerr has in mind [7] . In analogy with the metric sets of unital

C^{*}

-algebras, we define a metric set of an operator system

A

, in a strict operator system analogue (see Proposition 2.5 and Proposition 4.3 in [7] ), as a subset

K

A

which is norm compact, self-adjoint, and balanced, and convex, and separates the states on

A

. Then we have:

Corollary 5.5. Let

A

be an operator system and

K

a metric set of

A

For any subset

S

A

the following conditions are equivalent:

(1) The set $S$ is norm relatively compact.
(2) The set of affine functions ${\hat{a} \in A (S (A)) : a \in S}$ is bounded and equicontinuous with respect to the $w^{*}$ -topology on the state space $S (A)$ .
(3) The set $S$ is bounded and for every $ε > 0$ there exists a $λ > 0$ such that $S \subseteq A_{ε} + λ K + C 1,$ where $A_{ε} = {a \in A : ∥ a ∥ \leq ε}$ .

Proof. Since

A

is complete,

K

is weakly closed and absolutely convex. So there exists a weakly closed absolutely matrix convex set

K = (K_{n})

with

K_{1} = K

and

K_{n} \subseteq M_{n} (A)

for

n \in N

(see page 181 in [6] ). It is easy to verify that

K

is a matrix metric set of

A

when view

(A, 1)

as a matrix order unit space.

Fix

k_{n} \in K_{n}

for

n \in N

. Let

S = (S_{n})

with

S_{1} = S

and

S_{n} = {k_{n}}

for

n > 1

. By Theorem 5.4 , the following conditions are equivalent:

(i) The graded set

S

is norm relatively compact. (ii) The graded set

\hat{S} = ({\hat{S}}_{n})

, where

{\hat{S}}_{n} = {\hat{a} \in A (C S (A), M_{n}) : a \in S_{n}}

, is bounded and equicontinuous with respect to the matrix metric

D_{K}

and the matrix metric

ℰ^{(n)}

induced by the matrix norm on each

M_{n}

(iii) The graded set

S

is bounded and for every

ε > 0

there exists a sequence

λ = (λ_{n})

with

λ_{n} > 0

such that

S \subseteq A_{ε} + λ K + C 1 .

Clearly,

S

is norm relatively compact if and only if

S

is norm relatively compact.

It is also obvious that (iii) holds exactly if

S

is bounded and for every

ε > 0

there exists a

λ > 0

such that

S \subseteq A_{ε} + λ K + C 1 .

Now we need only show that the conditions (2) and (ii) are equivalent. Assume (2). Then

{\hat{a} \in A (S (A)) : a \in S}

is bounded. So there is an

M > 0

such that

q_{1} (\hat{a}) = sup {| \hat{a} (φ) | : φ \in S (A)} \leq M,

for

a \in S

. By Lemma 5.3 and Lemma 5.2 , we have

p_{1} (\hat{a}) \leq 4 q_{1} (\hat{a}) \leq 4 M,

for

a \in S

. Thus

{\hat{S}}_{1} = {\hat{a} \in A (C S (A), C) : a \in S_{1}}

is bounded. For

n > 1

, it is clear that

{\hat{S}}_{n} = {{\hat{k}}_{n}}

is bounded by Lemma 5.2 .

From Proposition 4.3 and Corollary 4.4 , the

D_{K_{1}}

-topology on

S (A)

agrees with the

w^{*}

-topology. Since

{\hat{a} \in A (S (A)) : a \in S}

is equicontinuous with respect to the

w^{*}

-topology on

S (A)

{\hat{a} \in A (S (A)) : a \in S}

is equicontinuous with respect to the

D_{K_{1}}

-topology. Given

ε > 0

. For any

φ \in S (A)

, we can find

δ^{'} = δ (ε, φ, n) > 0

such that

| \hat{a} (φ) - \hat{a} (ψ) | < \frac{ε}{4 n^{2}},

for all

a \in S

and

ψ \in S (A)

with

D_{k_{1}} (φ, ψ) < δ^{'}

. That

S (A)

is compact in the

D_{K_{1}}

-topology means that there are

φ_{1}, \dots, φ_{t} \in S (A)

such that

S (A) \subseteq \cup_{i = 1}^{t} U (φ_{i}; \frac{δ_{i}}{2})

, where

δ_{i} = δ (ε, φ_{i}, n)

. Take

δ_{0} = \frac{1}{2} min {δ_{1}, \dots, δ_{t}}

. Then for

φ, ψ \in S (A)

with

D_{k_{1}} (φ, ψ) < δ_{0}

, there is a

φ_{i}

with

φ \in U (φ_{i}; \frac{δ_{i}}{2})

. So

D_{k_{1}} (ψ, φ_{i}) \leq D_{k_{1}} (ψ, φ) + D_{k_{1}} (φ, φ_{i}) < δ_{i}

. Hence for any

a \in S

we have

| \hat{a} (φ) - \hat{a} (ψ) | \leq | \hat{a} (φ) - \hat{a} (φ_{i}) | + | \hat{a} (φ_{i}) - \hat{a} (ψ) | < \frac{ε}{2 n^{2}} .

So there is a

δ = δ (ε, n) > 0

such that

| \hat{a} (φ) - \hat{a} (ψ) | < \frac{ε}{2 n^{2}},

for

a \in S

and

φ, ψ \in S (A)

with

D_{K_{1}} (φ, ψ) < δ

By Corollary 4.5 , there is a lower semicontinuous matrix Lip-gauge

ℒ = (L_{n})

(A, 1)

such that

D_{K_{k}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : L_{r} (a) \leq 1, r \in N},

for

φ, ψ \in C S_{k} (A)

and

k \in N

. From Proposition 3.3 in [16] , we have

D_{K_{k}} (φ, ψ) = sup {∥ ≪ φ, a ≫ - ≪ ψ, a ≫ ∥ : a = a^{*}, L_{r} (a) \leq 1, r \in N},

for

φ, ψ \in C S_{k} (A)

and

k \in N

. For

n \in N

and

φ = [φ_{s t}], ψ = [ψ_{s t}] \in C S_{n} (A)

, there are

φ_{s t}^{(j)} \in S (A)

j = 1, 2, 3, 4

and

s, t = 1, 2, \dots, n

, such that

φ_{s t} - ψ_{s t} = φ_{s t}^{(1)} - φ_{s t}^{(2)} + i (φ_{s t}^{(3)} - φ_{s t}^{(4)})

by Lemma 4.1 in [16] . For

b = b^{*} \in M_{r} (A)

with

L_{r} (b) \leq 1

, we have

\begin{matrix} ‖ ≪ φ_{s t}^{(1)}, b ≫ - ≪ φ_{s t}^{(2)}, b ≫ ‖ & \leq & ∥ ≪ φ_{s t}, b ≫ - ≪ ψ_{s t}, b ≫ ∥ \end{matrix}

\begin{matrix} = & ∥ (e_{s} \otimes 1_{r}) (≪ φ, b ≫ - ≪ ψ, b ≫) (e_{t}^{*} \otimes 1_{r}) ∥ \end{matrix}

\begin{matrix} \leq & ∥ ≪ φ, b ≫ - ≪ ψ, b ≫ ∥ \end{matrix}

\begin{matrix} \leq & D_{K_{n}} (φ, ψ), \end{matrix}

where

e_{s} = [0 \dots 01_{s} 0 \dots 0]

. So we have

D_{K_{1}} (φ_{s t}^{(1)}, φ_{s t}^{(2)}) \leq D_{K_{n}} (φ, ψ)

for

s, t = 1, 2, \dots, n

. Similarly we have

D_{K_{1}} (φ_{s t}^{(3)}, φ_{s t}^{(4)}) \leq D_{K_{n}} (φ, ψ)

for

s, t = 1, 2, \dots, n

When

D_{K_{n}} (φ, ψ) < δ

, we have

D_{K_{1}} (φ_{s t}^{(1)}, φ_{s t}^{(2)}) < δ

and

D_{K_{1}} (φ_{s t}^{(3)}, φ_{s t}^{(4)}) < δ

, and hence

\begin{matrix} E_{n}^{(1)} ({\hat{a}}_{n} (φ), {\hat{a}}_{n} (ψ)) & = & ∥ φ (a) - ψ (a) ∥ \end{matrix}

\begin{matrix} \leq & \sum_{s, t = 1}^{n} | φ_{s t} (a) - ψ_{s t} (a) | \end{matrix}

\begin{matrix} \leq & \sum_{s, t = 1}^{n} (| φ_{s t}^{(1)} (a) - φ_{s t}^{(2)} (a) | + | φ_{s t}^{(3)} (a) + φ_{s t}^{(4)} (a) |) < ε, \end{matrix}

for

a \in S

. Hence

{\hat{S}}_{1}

is equicontinuous with respect to

D_{K}

and

ℰ^{(1)}

. For

n > 1

{\hat{S}}_{n} = {{\hat{k}}_{n}}

is equicontinuous with respect to

D_{K}

and

ℰ^{(n)}

since

{\hat{k}}_{n} \in C (C S (A), M_{n})

, whence (ii). The implication (ii) implies (2) is obvious. So conditions (2) and (ii) are equivalent. □

Remark 5.6. (1) According to Definition 3.1 in [1] , a metric set in a unital

C^{*}

-algebra may be not self-adjoint. For example, let

G = {e, g_{1}, g_{1}^{- 1}, g_{2}, g_{3}, \dots}

be a countable group, where

e

is the identity element of

G

and

g_{1}^{2} \neq e

. We denote by

K

the closed convex hull of the set

S = {δ λ_{e} : δ \in C, | δ | \leq 1} \cup {θ λ_{g_{1}} : θ \in C, | θ | \leq \frac{1}{2}} \cup {α λ_{g_{n}} + β λ_{g_{n}}^{*} : | α | + | β | \leq \frac{1}{n + 1}, α, β \in C, n \geq 2, n \in N}

. Then

K

is norm compact, balanced and convex, and separates the states on

C_{r}^{*} (G)

. But

K

is not self-adjoint.

(2) We would like to point out that the definition of a metric set in [1] should contain the condition of self-adjointness. One reason is that a Lipschitz seminorm

L

for a

C^{*}

-algebra

A

should satisfy

L (a^{*}) = L (a)

for

a \in A

(see page 6 in [12] or Proposition 2.5 and Proposition 4.3 in [7] ). Another reason is that without the self-adjointness, we can not get

\forall h \in ℋ \forall γ \in A_{2}^{*} \cap {C I}^{⊥} \cap δ (K^{\circ}) : | γ (h) | \leq 2 ε

from

\forall h \in ℋ \forall γ \in (A_{h}^{*})_{2} \cap {C I}^{⊥} \cap δ (K^{\circ}) : | γ (h) | \leq ε,

(see page 258 in [1] ) because it is not guaranteed that

\frac{1}{2} (γ + γ^{*})

and

\frac{1}{2} (γ - γ^{*})

belong to

δ (K^{\circ})

for

γ \in δ (K^{\circ})

Acknowledgements

I would like to thank Marc Rieffel for valuable discussions and suggestions. This research was partially supported by Shanghai Priority Academic Discipline, China Scholarship Council and National Natural Science Foundation of China. References

C. Antonescu, E. Christensen, Metrics on group $C^{*}$ -algebras and a non-commutative Arzelà-Ascoli theorem, J. Funct. Anal., 214 (2004), 247–259.
W. B. Arveson, Subalgebras of $C^{*}$ -algebras, Acta Math., 123 (1969), 141–224.
M. -D. Choi, E. G. Effros, Injectivity and operator spaces, J. Funct. Anal., 24 (1977), 156–209.
A. Connes, Compact metric spaces, Fredholm modules, and hyperfiniteness, Ergodic Theory Dynam. Systems,9(2) (1989), 207–220.
E. G. Effros, Advances in quantized functional analysis, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif., 1986), 906–916, Amer. Math. Soc., Providence, RI, 1987.
E. G. Effros and C. Webster, Operator analogues of locally convex spaces, Operator algebras and applications(Samos, 1996), 163–207, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 495, Kluwer Acad. Publ., Dordrecht, 1997.
D. Kerr, Matricial quantum Gromov-Hausdorff distance. J. Funct. Anal., 205 (2003), 132–167.
V. I. Paulsen, Completely bounded maps and dilations, Pitman Research Notes in Mathematics Series, 146. Longman Scientific & Technical, Harlow; John Wiley & Sons, Inc., New York, 1986.
M. A. Rieffel, Metrics on states from actions of compact groups, Doc. Math., 3 (1998), 215–229.
M. A. Rieffel, Metrics on state spaces, Doc. Math., 4 (1999), 559–600.
M. A. Rieffel, Group $C^{*}$ -algebras as compact quantum metric spaces, Doc. Math., 7 (2002), 605–651.
M. A. Rieffel, Gromov-Hausdorff distance for quantum metric spaces, Mem. Amer. Math. Soc. 168 (2004), 1–65.
M. A. Rieffel, Matrix algebras converge to the sphere for quantum Gromov-Hausdorff distance, Mem. Amer. Math. Soc. 168 (2004), 67–91.
C. Webster and S. Winkler, The Krein-Milman theorem in operator convexity, Trans. Amer. Math. Soc., 351 (1999), 307–322.
W. Wu, Non-commutative metric topology on matrix state spaces, to appear in Proc. Amer. Math. Soc., arXiv:math.OA/0410587.
W. Wu, Non-commutative metrics on matrix state spaces, arXiv:math.OA/0411475.
W. Wu, Quantized Gromov-Hausdorff distance, arXiv:math.OA/0503344.

Department of Mathematics, East China Normal University, Shanghai 200062, P.R. China E-mail address : wwu@math.ecnu.edu.cn Current address : Department of Mathematics, University of California, Berkeley, CA 94720-3840 E-mail address : wwu@math.berkeley.edu

wei wu